Run Milvus, the open-source vector database behind many RAG pipelines and semantic search engines, from a pre-configured image in your own cloud account. You pay for the server, not per stored vector or per query.
Milvus is an open-source database built for storing, indexing and searching embedding vectors, the numbers that text, images, audio and video are turned into before an AI model can compare them. Instead of matching keywords, it finds the nearest vectors to a query, which is what powers retrieval-augmented generation (RAG), semantic search and recommendations.
Store document embeddings and retrieve the closest matches to feed your LLM.
Search by meaning rather than exact keywords across text, images and more.
Find similar products, content or users from their vector representations.
Match images and video frames against large collections of embeddings.
Connect with pymilvus or any client that speaks the Milvus protocol on port 19530.
No per-vector or per-query billing, and your data stays in your own account.
Running Milvus by hand means Docker Compose files, dependency versions and several supporting services to wire together. The marketplace image arrives with all of that done.
Subscribe on AWS Marketplace or launch from Google Cloud Marketplace, pick a region and machine size, and deploy. Allow a few minutes for Milvus to start.
The AWS security group opens it for you. On Google Cloud, confirm a firewall rule allows TCP 19530 from the addresses that will connect.
Point pymilvus at the server's public IP on port 19530 and sign in with the default root user.
Change the default root password before storing real data, and limit the allowed source IPs instead of leaving the port open to everyone.
An AMI on AWS Marketplace, launched into your own AWS account with one-click launch or from the EC2 console.
Launch on AWS 
The same database as a Google Cloud Marketplace image, deployed into your own GCP project.
Launch on Google Cloud Allow these ports in your AWS security group or GCP firewall. If pymilvus cannot connect, check port 19530 first.
| Service | Ports | Protocol |
|---|---|---|
| Milvus client / gRPC | 19530 | TCP |
| HTTP and HTTPS | 80, 443 | TCP |
| SSH (administration) | 22 | TCP |
Server size: t3a.large on AWS or e2-standard-2 (2 vCPU, 8 GB RAM) on Google Cloud to start. Choose a larger machine as your collections and query volume grow.
| Milvus by Meetrix | Pinecone | Zilliz Cloud | |
|---|---|---|---|
| Hosting | Your AWS account or GCP project | Pinecone's cloud, no self-host option | Zilliz's managed cloud, or a self-hosted enterprise tier |
| Data control | Vectors never leave your account | Pinecone stores and indexes your vectors | Zilliz stores and indexes your vectors on the hosted tier |
| Pricing model | Cloud compute only, no per-query fee | Usage-based, per stored vector and per query | Usage-based, per compute unit on the hosted tier |
| Data residency | Any region you choose | Limited region choice, Pinecone's terms apply | Limited region choice, Zilliz's terms apply |
Zilliz Cloud is the managed version of Milvus. Read the full Milvus on AWS comparison →
Storing and searching embedding vectors at scale: retrieval-augmented generation (RAG), semantic search, recommendation engines, and image or video similarity search. If your app turns content into vectors and needs the closest matches fast, that is what Milvus does.
Milvus itself is open source and free. Through the Meetrix listings you pay standard AWS or Google Cloud compute costs for the server, plus any Meetrix software fee shown on the marketplace listing, not a per-vector or per-query fee.
t3a.large is the recommended starting point on AWS, and e2-standard-2 (2 vCPU, 8 GB RAM) with a 20 GB disk on Google Cloud. Resize the instance as your collections and query volume grow.
Use the pymilvus SDK with the server's public IP, port 19530 and the default root user. Visiting http://<server IP> in a browser shows a confirmation page once Milvus is running.
The image ships with the admin user root and the password Milvus. Change the password before storing anything beyond test data.
No. Milvus does not ship one and the image stays lean. If you want a UI, run Attu, the open-source Milvus management tool, in a separate container pointed at port 19530.
Yes. It is a standard Milvus instance, so in most cases you only change the host and port your client points at.
When a new version is published on the marketplace, back up your Milvus data, remove the old deployment and launch the new version.
Run your own vector database without the per-query bill, and how it compares to Pinecone and Zilliz Cloud.
Deploy Milvus on Google Cloud in minutes and keep embeddings in the region you choose.
Configure the deployment, open port 19530 and connect with pymilvus, step by step.
We build and run self-hosted AI infrastructure for teams every day, from instance sizing to index tuning. Tell us what you need.
Contact us